Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Dynamic Panel Data

From   Clive Nicholas <>
Subject   Re: st: Dynamic Panel Data
Date   Sun, 6 Mar 2011 20:44:27 +0000

Eric De Souza replied to Humaira Asad:

> First, let's ignore the fact that your data are five year averages, and consider them as successive time periods. In this case, the introduction of a lagged value of the dependent variable as regressor makes FE, RE and FD inconsistent because the lagged dependent variable, when transformed in order to apply the above methods are correlated with the residuals. You, therefore, have to use instrumental variables. LDV, because it is identical in effects to FE, has the same problem. Stata has a dynamic panel data routine but I have never used it. What I use is -xtabond2- written by David Roodman and which can be downloaded using -ssc install xtabond2- . The method is not simple and would require some reading up on your part. But if you Google, you will find quite a lot of course notes explaining it. You also have the paper by David Roodman himself, but I personally do not recommend it for beginners. If you wish to use the built-in Stata commands , they are -xtabond- and -xtdpds!


Your position is, of course, supported by Judson and Owen (1999). What
if, however, the coefficient on the LDV is insignificant? Could not
the problem then be ignored? In any case, could one not get round the
problem entirely by using -xtpcse- with the -c(psar1)- option? Nat
Beck and Jon Katz, in a 2004 paper, certainly think so. They write, in
the abstract:

"It is shown that there is nothing pernicious in using a lagged
dependent variable, and all dynamic models either implicitly or
explicitly have such a variable; the differences between the models
relate to assumptions about the speeds of adjustment of measured and
unmeasured variables. ... It is noted that models with both a lagged
dependent variable and serially correlated errors can easily be
estimated; it is only OLS that is inconsistent in this situation. We
then show, via Monte Carlo analysis shows that for typical TSCS data
that fixed effects with a lagged dependent variable performs about
as well as the much more complicated Kiviet estimator, and better than
the Anderson-Hsiao estimator (both designed for panels)."

See for the full
paper. I've mentioned Beck, Katz and -xtpcse- so many times in
response to Statalist queries over the years, people must think I'm on
commission, but I'm not a cab for hire just yet.

Clive Nicholas

[Please DO NOT mail me personally here, but at
<>. Please respond to contributions I make in
a list thread here. Thanks!]

Judson KA and Owen AL (1999) "Estimating Dynamic Panel Data Models: A
Guide for Macroeconomists", Economics Letters (65): 9-15.

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index